Izvorni znanstveni članak

https://doi.org/10.21278/TOF.501075124

Dynamic Reconfiguration of UAV Formations Using RL-Enhanced Swarm Control

Vasantharaj Rajagopal ; Department of Aerospace Engineering, MIT Campus, Anna University, Chennai-600044, Tamil Nadu, India
K. Senthil Kumar ; Department of Aerospace Engineering, MIT Campus, Anna University, Chennai-600044, Tamil Nadu, India *

* Dopisni autor.

Sažetak

Swarms of Unmanned Aerial Vehicles (UAVs) have been widely used in various applications mainly for surveillance and crowd-sensing. A swarm setup is an environment where multiple UAVs coordinate together to execute a specific mission. Such a swarm of UAVs is a useful entity to access areas where human penetration is impossible. To facilitate the need for automation among swarms of UAVs, a Deep Q-Learning based Dynamic Swarm Pattern Formation (DSPF) model is proposed. The proposed Speed Control based Reinforcement Learning (SC-RL) algorithm enhances the DSPF model to achieve pattern formation in an automated manner. The SC-RL algorithm strives to switch patterns efficiently by avoiding inter-UAV collisions and also keeps optimal trajectory intact throughout its pattern switching mechanism. To speed up the pattern building process and enable parallelised coordinate computation, a Decentralised Coordinate Computation (DCC) algorithm is implemented. The Servo Interrupt based Pattern Switch (SIPS) control also gives the DSPF model the ability to change patterns dynamically, which allows it to be adjusted to a variety of situations. By increasing the pattern formation time and distance covered by about 95.68% and 66.67%, respectively, simulations conducted for 100 UAVs demonstrate the viability of the suggested DSPF model in a crowded, collision-prone environment.

Ključne riječi

deep learning; multi-agent system; pattern formation; reinforcement learning; simulation; SWARM

Hrčak ID:

346013

URI

https://hrcak.srce.hr/346013

Datum izdavanja:

22.1.2026.

Posjeta: 275 *